Meta-4mCpred

Meta-4mCpred predicts N4-methylcytosine (4mC) sites in DNA sequences to support analysis of DNA methylation and epigenetic regulation.

Key Features:

• Meta-predictor framework: Integrates multiple machine learning algorithms into a single meta-predictor to improve prediction accuracy and generalizability across species.
• Feature representation learning scheme: Generates 56 probabilistic features derived from four distinct machine-learning algorithms that capture compositional, physicochemical, and position-specific sequence attributes.
• Support vector machine integration: Uses the 56 probabilistic features as inputs to a support vector machine forming the final meta-predictor model.
• Seven feature encodings: Incorporates seven different feature encodings to capture comprehensive sequence information.
• Cross-species validation: Achieves an overall average accuracy of 84.2% via cross-validation, representing an approximate 2%–4% improvement over existing predictors.
• Independent dataset evaluation: Attains an overall average accuracy of 86% on independent datasets, surpassing current leading predictors by more than 4%.

Scientific Applications:

• DNA methylation mapping: Predicts 4mC site distribution to facilitate studies of N4-methylcytosine patterns in genomic DNA.
• Epigenetic regulation research: Supports investigation of methylation-mediated regulation of gene expression, DNA replication, and cellular differentiation.
• Comparative genomics and evolutionary analysis: Enables cross-species analyses of 4mC distribution for evolutionary and comparative studies.

Methodology:

Four machine-learning algorithms were used to extract 56 probabilistic features across seven feature encodings, and those features were integrated by a support vector machine into the final meta-predictor.